Internet Link Delay Estimation's Study

Internet is increasing larger and more complex. It is important for the society that the whole network is stable. To better design, control and manage dynamic networks, it is essential to have link-level performance data for our understanding internal network characteristics, such as link loss and delay. Data collecting from the edge of the network is the only data based on end to end measurement. Instead of classical statistics approach, here several models and algorithms are proposed to efficiently and accurately assume the internal data in networks.Various network applications in the network, lead to different traffic load, and unbalanced impact on each link. Based on end-to-end measurement, called as network tomography problem, was prompted by several international organizations.This essay emphasizes on the following researches about link-level's delay estimation.1. A link delay distribution approach is proposed based on the dependence tree model and upward-downward algorithm. Comparing with classical maximum likelihood estimation (MLE), the main difference is the posterior probability is assumed based on the prior probability on the same link. This process is repeated to generate estimated numbers to adapt traffic changes. Furthermore, according to a special dependence tree model, multicast dependence tree model, the upward-downward algorithm was optimized to reduce to one estimated parameter. The simulation results of the algorithm prove the accuracy of the upward-downward algorithm; moreover, derive the delay characteristic efficiently based on the end-to-end measurement data. Simulation of the algorithm was implemented using C/C++.2. Based on Mixed Finite Mixture Models (MFMM). with Expectation...